]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
selftests: drv-net: gro: increase the rcvbuf size
authorJakub Kicinski <kuba@kernel.org>
Wed, 7 Jan 2026 23:25:57 +0000 (15:25 -0800)
committerJakub Kicinski <kuba@kernel.org>
Sat, 10 Jan 2026 01:34:30 +0000 (17:34 -0800)
commita0ac0ff382767a5dbde266fb5f7997a5d6b70e10
treedfb9d6b0788c60aaa16bf21370f7447e0af11eca
parent96ea4fa60c4528d95bdbce7f4212c015ab3e8113
selftests: drv-net: gro: increase the rcvbuf size

The gro.py test (testing software GRO) is slightly flaky when
running against fbnic. We see one flake per roughly 20 runs in NIPA,
mostly in ipip.large, and always including some EAGAIN:

  # Shouldn't coalesce if exceed IP max pkt size: Test succeeded
  # Expected {65475 899 }, Total 2 packets
  # Received {65475 899 }, Total 2 packets.
  # Expected {64576 900 900 }, Total 3 packets
  # Received {64576 /home/virtme/testing/wt-24/tools/testing/selftests/drivers/net/gro: could not receive: Resource temporarily unavailable

The test sends 2 large frames (64k + change). Looks like the default
packet socket rcvbuf (~200kB) may not be large enough to hold them.
Bump the rcvbuf to 1MB.

Add a debug print showing socket statistics to make debugging this
issue easier in the future. Without the rcvbuf increase we see:

  # Shouldn't coalesce if exceed IP max pkt size: Test succeeded
  # Expected {65475 899 }, Total 2 packets
  # Received {65475 899 }, Total 2 packets.
  # Expected {64576 900 900 }, Total 3 packets
  # Received {64576 Socket stats: packets=7, drops=3
                    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  # /home/virtme/testing/wt-24/tools/testing/selftests/drivers/net/gro: could not receive: Resource temporarily unavailable

Reviewed-by: Willem de Bruijn <willemb@google.com>
Link: https://patch.msgid.link/20260107232557.2147760-1-kuba@kernel.org
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
tools/testing/selftests/drivers/net/gro.c